Retail in particular, both online and brick-and-mortar, has great potential and offers a wide range of application areas for generating competitive advantages through data-based decisions. E-commerce in particular is predestined to implement innovations in this area due to the naturally high degree of digitization and the large amount of available data. This is not only reserved for large retail companies and e-commerce platforms, but smaller retailers in particular can use their data treasures to add value with the help of AI.

A prominent use case is forecasting: Whether sales planning, incoming goods or returns forecasting, a good forecast provides the basis for any planning. Unfortunately, previous planning tools, such as the frequently used Excel spreadsheet, the well-known gut feeling or even conventional planning software, quickly reach their limits with the increasing complexity and amount of available data. Here, the use of machine learning algorithms makes it possible to take into account many different data sources and considerable quantities, so that more consistent, accurate and transparent forecasts can be created, ultimately increasing planning reliability. We will show you what this means for typical forecasting use cases in retail using five examples from the world of online and stationary retail.

WHAT IS FORECASTING?

Forecasting is about predicting future events as accurately as possible based on historical observations and data, as well as knowledge of future events that could affect the forecast.

In this regard, machine learning algorithms are perfect for a wide variety of forecasting applications: They automatically learn patterns and relationships in historical data, which can then be applied to new data to make predictions. Given the appropriate data, a wide variety of things can be predicted in this way, as the following examples show.

EXAMPLE #1: FORECASTING SALES IN THE FOOD TRADE

What merchandise is needed in the individual sales outlets on a given day? A large supermarket can have up to 30,000 items in its assortment, which means that the supermarket needs as meaningful a forecast as possible for all 30,000 items. A particular difficulty here is the varying sales frequency of individual items. This is further complicated by the fact that many assortment items have varying and in some cases extremely short sales histories, such as newly introduced items.

In food retailing in particular, machine learning-based demand forecasting has great potential, not only for retailers' purchasing and inventory planning, but also in terms of sustainability. By more accurately forecasting the purchasing behavior of end consumers per day and per store, the purchasing manager knows how many items to put on the shelf to avoid empty shelves, full warehouses, or throwaways of perishable items. According to a recent study by Bitkom and BVE, intelligent sales forecasts could reduce food waste to zero by 2030 ("Zero Waste").

On our blog, you can read the joint success story with a supermarket chain about how we succeeded in reliably predicting the sales of racers and bums.

Free whitepaper on sales forecasting

The sales forecast often forms the basis for further forecasts on inventory levels, personnel requirements and returns. Learn more about the advantages of machine learning-based forecasts in our free whitepaper.

Download

EXAMPLE #2: FORECASTING INVENTORIES IN THE TEXTILE RETAIL TRADE

The main goal of inventory optimization is to keep stock levels as low as possible while still being able to offer customers the right product at the right place at the right time. With ever-changing customer demand and in the face of short product life cycles and the ever-increasing prevalence of omni-channel models, this is no easy task. For example, according to a study by IHL Group, in retail alone, out-of-stocks cause $634 billion in lost global sales each year, while overstocks due to markdowns result in $472 billion in lost sales. As a core part of a fashion house's buying department, a forecast can ensure that these exact scenarios are avoided.

For example, historical sales figures along with individual product price trends can be analyzed using machine learning algorithms to forecast future product demand. Taking current stock levels into account, the best possible order quantity can then be recommended directly to ensure that there is neither too much nor too little product in stock to meet the predicted demand. The automated creation of the individual order quantity forecast represents a significant workload reduction in purchasing planning. The newly gained time allows the fashion house's buyers to invest more effort in evaluating the merchandise again in order to guarantee the best possible quality.

EXAMPLE #3: FORECASTING INCOMING GOODS QUANTITIES IN RETAIL LOGISTICS

A logistics service provider for a large German fashion house wants to know how incoming goods are behaving in the warehouses in order to enable more precise personnel planning. Especially in the fashion industry, delivery dates and goods receipts are often difficult to plan. Instead of exact delivery dates, approximate time frames are prevalent. If delivery dates are not met and suppliers deliver their goods seemingly at random, it is almost impossible to optimally plan goods receipts and thus personnel and warehouse utilization. There are regularly too many or too few warehouse staff, resulting in high costs and productivity losses.

Machine learning can help here as well, accurately predicting goods receipts from different suppliers. Thus, a combination of supplier evaluation and Machine Learning can be used to create a concrete prediction for the expected time of delivery of the respective supplier's goods. This not only enables more efficient planning of personnel, but also maximum utilization of the warehouse.

EXAMPLE #4: THE FORECAST OF VISITORS IN THE RETAIL TRADE

Fluctuating visitor numbers in the retail sector make reliable staff planning difficult. Particularly for stores that are characterized by products with a high level of consultation and support, it is important to know how many customers can be expected on any given day and over the course of the day. To ensure optimum customer care, an optician therefore wants to know how many customers can be expected over the course of the hour, day and week, so that he can plan the staff required at the individual stores accordingly. Thanks to frequency counters at the entrance to the stores, information about the flow of customers is already available. The daily customer flow is made up of walk-in customers and customers with agreed appointments.

Based on historical visitor numbers and with the help of machine learning algorithms, a reliable forecast of customer traffic at each of the optician's stores can be generated and a recommendation of staffing requirements can be made automatically on the basis of the calculated forecast. When training the machine learning models, a large number of other influencing factors, such as the weather and information on special opening times or shortened working weeks and vacations, can be included and checked for their effect on the forecast. With the help of the forecasts, staff planning and customer service can be optimized, costs reduced, and direct added value generated for customers at the same time.

EXAMPLE #5: FORECASTING RETURN QUANTITIES IN FASHION ECOMMERCE

Despite all efforts, returns are simply part and parcel of online retailing. This is especially true in the fashion industry. However, returns processing is a time-consuming and cost-intensive process. In order to keep the costs for returns processing in check and to speed up returns handling, a logistics service provider responsible for the online store of a fashion retailer therefore wants to know how many returned packages are to be expected today, tomorrow and in the next few weeks.

Here, too, machine learning comes to the rescue by predicting the number of returns. This makes it possible to forecast how many returned parcels can be expected and when. In this way, the daily volume of returns can be predicted and thus personnel deployment can be better planned and logistics resources can be managed in a targeted manner. The returns quantity forecast thus simplifies capacity planning for logistics, warehouses and staff in view of fluctuating parcel volumes.

Would you like to know more about how returns forecasting works? Then read our joint success story with Mode Logistik GmbH & Co. KG, the logistics operator of Fashion ID, the online store of Peek & Cloppenburg KG Düsseldorf.

SUMMARY

No matter whether B2B or B2C, whether stationary or online and no matter which product category is involved: The possible uses of forecasting are more than diverse and the examples mentioned naturally do not cover all fields of application by far. The examples are only meant to be a source of inspiration and to help you get an idea of what you too can forecast in your company with Machine Learning.

Are you interested? Then our forecasting experts will be happy to help you find out what your individual forecast could look like.

Free Whitepaper
Forecasting with Machine Learning

The sales forecast often forms the basis for further forecasts on inventory levels, personnel requirements and returns. Learn more about the advantages of machine learning-based forecasts in our free whitep

Download